Facebook's AI lab

https://www.facebook.com/yann.lecun/posts/10151728212367143

by vkhuc • 13 years ago

238 91 13 years ago

mturmon3• 13 years ago

Star power at NIPS:

"Facebook CEO Mark Zuckerberg, CTO Michael Schroepfer and I are at the Neural Information Processing Systems Conference in Lake Tahoe today. Mark will announce the news during his presentation at the NIPS Workshop on Deep Learning later today."

It seems Zuckerberg will have an apres-ski Q&A, and be in a panel discussion at day's end (https://sites.google.com/site/deeplearningworkshopnips2013/s...). Hope the room is big enough.

btown3• 13 years ago

At NIPS now. There have been rumors going around that Zuckerberg might have just been interested in attending, giving a casual Q&A, not announcing anything like this - but with this announcement, I foresee them having to take down all of the dividers in Harveys ;)

jekdoce2• 13 years ago

Could you post some bullet points of what was said there? Would be really appreciated by everyone I'm sure!

turing• 13 years ago

Wow. With Hinton at Google and LeCun at Facebook, industry has definitely been pulling some big names away from academia.

yeukhon2• 13 years ago

I beg to differ. They still publish papers and big names in academia are often consultants or receiving grants from private industry. And according to LeCun he will remain as professor at NYU and run his research lab from NYU. He will serve as director and lead researcher.

fat0wl3• 13 years ago

Oh man that guy was my machine learning professor at NYU :) Sparked my interested in Lisp, but now I seem to recall that he wrote all the problem examples in his own language.... Lush. Even so, all my knowledge of Lisp syntax and neural nets stems from his bad-ass lectures

seiji• 13 years ago

I beg to differ.

How many people are they teaching in person these days?

mturmon3• 13 years ago

You are mistaken.

LeCun is now running a high profile lab for a $100B company. He's got the personal attention of the founder. He's only going to be at NYU as a talent conduit. It's not the same as Hinton's arrangement at all.

zerooneinfinity3• 13 years ago

Really changes the phrase 'those who can't do teach'.

bch• 13 years ago

That sassy phrase (Those who can, do; those who can't, teach.) is irritating to me. Now I'm not casting judgement on anybody who _does_ find it entertaining, or useful, but for what it's worth, the _original_ phrase is:

Those who know, do; those who understand, teach.

-Aristotle.

1il7890• 13 years ago

Looks awesome for once!

ayubskhan• 13 years ago

Cool! Will soon see online/offline AI in action!

dhammack• 13 years ago

Can you imagine the datasets FB has to work with? Statuses, pictures, locations, social graphs. It's incredible. It's hard to blame LeCun/Hinton/Ng for moving towards industry with data like that. I'd bet that the only place with more data than FB/Google is our good friends at the NSA.

msoad• 13 years ago

My friend used to work at NSA research. He said they don't have much ready to use data yet

iLoch• 13 years ago

Nice try, NSA.

btown3• 13 years ago

Apparently, it's because NSA researchers are spending all their time playing WoW instead.

TrainedMonkey2• 13 years ago

This is huge. LeCun is one of the leading AI experts in the world. With resources of Facebook I expect great things. I also see this as a technology bet by Facebook on long term strategy.

I own Facebook stock, so I might be biased.

oatmealsnap• 13 years ago

Will "great things" be "better ads"? Or maybe something worth while...

swalsh• 13 years ago

The AI facebook has today is already freaky to me at times... while I love any new advancement in AI, these advancements will probably not be used in my interests.

mikeg8• 13 years ago

Completely agreed. Not looking that forward to this.

logn• 13 years ago

Companies like Facebook and Google do a lot though to promote these areas by building great generic tools. Their work in, say, map-reduce has been invaluable to a wide range of applications.

Irishsteve• 13 years ago

It looks like they don't have anything particularly taxing compared to the likes of Google. There is only so much they can apply to the FB feed. Graph search while cool still isn't that advanced.

Schwolop• 13 years ago

There's plenty of data on the world wide web that pertains to you but isn't linked to Facebook - yet. AI that can do context aware authorship analysis could have huge returns for Facebook.

TrainedMonkey2• 13 years ago

Value of Facebook consists of two things, you and your data. It is not good for long term strategy of Facebook to hurt either.

Here are some things where smart AI could help: 1. Better filtering of spam bots. 2. Better filtering of ads that would significantly improve ad quality. 3. Smart assistant that learns your habits and makes smart suggestions such as: which events to attend, your writing style, and filters immediate notifications based on personalized importance score.

Or Facebook could exploit machine learning/AI techniques to simply drive up click through ratio on their ads.

Tossrock• 13 years ago

The "which events to attend" thing is already up and running

stokedmartin• 13 years ago

3. Smart assistant that learns your habits and makes smart suggestions such as: ...

Recognize your romantic partner using FB network[0]

[0] https://news.ycombinator.com/item?id=6759866

dnautics• 13 years ago

although I agree - it might be worth considering it's creepy because of the uncanny valley

nl• 13 years ago

Uncanny Valley doesn't really happen with AI on the internet.

People are already used to people saying and doing dumb things online, so something that is nearly "smart human like" is a big advance (on average).

dundun2• 13 years ago

Yann deals with Machine Learning. His specialty (simplified) was pattern recognition in images (first handwriting then objects).

I'd imagine projects that group similar topics (i.e. these people are all in front of the Great Pyramids or these people are all taking pictures of their babies) which probably will help group/filter content

qznc• 13 years ago

"It should be noted that no ethically-trained software engineer would ever consent to write a DestroyBaghdad procedure. Basic professional ethics would instead require him to write a DestroyCity procedure, to which Baghdad could be given as a parameter." -- Nathaniel Borenstein

lucb1e3• 13 years ago

I'm not sure what I should think of this. What are your thoughts with regard to the amount of personal information that Facebook possesses? Regardless of whether they are good or evil, what do you think the potential implications are? I'm curious about HN's thoughts.

anoncowherd• 13 years ago

> What are your thoughts with regard to the amount of personal information that Facebook possesses?

"Fucking disgusting"

> Regardless of whether they are good or evil, what do you think the potential implications are?

Less privacy, more data in the hands of governments, more police state.

res0nat0r3• 13 years ago

The FB data is being given to them freely by their users

X4• 13 years ago

Is any of the conference videos available online, or is that only for paying customers?

ezioamf• 13 years ago

Google has won the search indexing game. Who will win the next level when robots understands the content?

demonicus• 13 years ago

"We were foolish to think Google was the entity that would become Skynet, when it was Facebook all along." - John Connor

larrybolt2• 13 years ago

AI always inspired me to become better at programming and learn new languages, methods and idea's. And even with the currently available tools and technologies, it amazes me how little of these dreams I had actually exist or are being worked on.

EDIT: Just came across an article: "The Mother of All Demos" is 45 today; Douglas Engelbart did exactly what I'm hoping someone today will be able to pull of.

xerophtye• 13 years ago

Link to the article?

loceng• 13 years ago

This is all about finding patterns for businesses to better target and sell to people. Valuable information is likely NOT going to reach or become usable by say government organizations or other organizations that are trying to problem solve big problems - at least not be affordable, because they'll be competing with for-profit businesses for bidding on access those users.

aioprisan• 13 years ago

I took LeCun's class in undergrad at NYU and it definitely wasn't easy. He's a fantastic speaker, definitely made it more enjoyable.

krebby• 13 years ago

Same here. His Machine Learning course was one of the hardest but most rewarding classes I've ever taken. Definitely glad to see him expanding his reach outside of Academia.

yangtheman• 13 years ago

Am I the only one who is skeptical about this announcement? I am sure the amount of data Facebook has will be huge asset to any sort of AI development, but precisely because of the amount of data and the kinds of data they have, it's just scary what this could be used for...

AI that has personal information of 500M+ people, using it to manipulate people...(first to click on advertisements, and then much more). In the hands of government, I shudder to think what's possible. With NSA already snooping around, perhaps it's not all that distant.

justaman3• 13 years ago

The announcement of an organization who's revenue is based on advertising doing research into AI deeply scares me. Take a look at this MIT study. Each day our corporation based government takes a step closer to 1984 and each day people become more and more apathetic and quick to change the subject at the first sign of the conversation.

http://web.mit.edu/people/amliu/Papers/PentlandLiu_NeuralCom...

tehf0x• 13 years ago

Umm what do you think google's revenue is based on?

eli_gottlieb• 13 years ago

That was published in 1999 and we still don't have pre-crime reporting security cameras flagging me everywhere I go.

Just saying.

jfoster• 13 years ago

Would facebook's data actually be useful to the NSA? I think they would get a lot more actionable data out of emails, SMSes, phone transcripts, accounting software, project management software, etc. The only thing I can imagine facebook being good for is perhaps knowing who they want to start looking at in the future. Search engine history might be more effective for that, though.

justaman3• 13 years ago

The point is, this data exists. The NSA has all of it. What are they doing with it? Probably nothing good.

jnardiello• 13 years ago

This is awesome and incredibly creepy at the same time.

DanBC2• 13 years ago

When Google announces AI research it makes sense because self driving cars and trying to understand what a user means by typing stuff in a search box.

But Facebook? They could create a million virtual DanBCs and then A/B test something against them, and then present me with the irresistible ad, perfectly pitched to draw me in?

derefr• 13 years ago

"X Labs"-type sub-divisions rarely do things of immediate practical interest to X.

Microsoft Research, for example, doesn't restrict itself to the domain of features for Windows, Office, or XBox. Instead, you get giant touchscreens (which you could say inspired Windows 8's unified tablet-PC approach, but didn't really) and PhotoDNA.

Likewise, I'd expect the things coming out of a Facebook AI Lab to be more of the "new ideas that could replace our entire business model when this one runs dry" than "better ways to advertise." Personal, private social agents that live in the cloud and continuously execute on your goals without requiring direct instruction, say (I'm sure there's a speculative-fiction name for these, but all that comes to mind is "NetNavis.")

gress• 13 years ago

Perhaps they want to understand what users mean too.

nacs• 13 years ago

And what makes you think Google can't or won't do the same? Their latest aggressive push of Google+ everywhere including the forced Youtube integration indicates that they have the capability or have been doing exactly what you suggest.

Not to mention the huge amounts of data they collect through searches and Gmail that they already have been using to tailor ads to each visitor.

DanBC2• 13 years ago

Oh, I totally agree.

I see Google announcing AI research, and I think "self driving car" or "understanding search terms", but I have no doubt that it's also "serve better ads" or "slurp more data".

You're right, I should have put that in.

zerooneinfinity3• 13 years ago

Seeing Peter Norvig congratulate him in the status gave me chills. Great times to be living around these titans indeed.

bborud• 13 years ago

In the eyes of Facebook, we are all undressed. Now it will be like they are throwing in a free colonoscopy while they're at it.

I bet the guys over at Fort Meade are beside themselves with happiness over this :-)

superfx• 13 years ago

Rob Fergus is joining Facebook too:

https://plus.google.com/103199870243230279182/posts/JaqgU1sX...

sureshv• 13 years ago

I hope these labs are as productive as Google or the old DEC labs: http://www.hpl.hp.com/techreports/Compaq-DEC/

achompas2• 13 years ago

random thoughts on this announcement

I took Yann's ML class during the first semester of my program at NYU two years ago. It was a terrible experience, given that I hadn't worked with linear algebra or mathematical statistics since about 2008. I could barely write Python at the time, so my Lush code was some ungodly non-functional crap. That said, I'm very happy I took the class in 2011 despite hyperventilating my way through the final--it was only offered once more, and I suspect it will never be offered again given today's news.

Yann really knows his stuff. His convolutional nets (and their application to MNIST and image segment classification) represented a significant improvement in computer vision, and he demoed some incredible low-latency image segmentation stuff for us. He runs one of the best neural net labs in the world (up there with Hinton's and Bengio's), and he has some incredible students at NYU. I can see this shaping up as a delayed acquihire of sorts...he will not struggle to find excellent candidates.

There's a lot of discussion around the privacy implications, but I think everyone's rehashing old points. Facebook already hires excellent researchers and data scientists -- John Myles White and Sean Taylor just moved out west -- but they don't focus on images just yet, from what I know. This hiring represents an investment in image analysis on Facebook's behalf that matches what they put into unstructured textual data and graphical inference. If you've already stayed with Facebook through the "graph search" announcement, this shouldn't surprise you either.

As someone interested in this type of work, it's an exciting time to live in New York City. Finance and adtech have been here forever, but things have expanded. Many startups (Foursquare, Tumblr, Knewton, Etsy) have invested in applied statistics and machine learning, hiring excellent researchers and engineers. Columbia and NYU have announced data science initiatives in the last 6 months. Very smart people (and others, like me ;) ) are very active in the community here.

There are some obvious applications of Yann's work to Facebook's advertising goals:

Identifying strong friendships through co-occurrence in photos
Digit or character recognition applied to marketing in photos (shop signs, brands, etc)
Image segment classification (e.g. beach, park, road) for use in predicting a photo's location (for those uploaded after the fact)
All the Instagram photos. I mean seriously. They're committed to making ads seamless--why not use image segmentation + likes to identify photographic structures people are attracted to?

Can anyone think of others?

pekk• 13 years ago

Facebook has a known, established history of carrying out large-scale, intrusive personal surveillance. Imagine if I said "NSA already hires excellent researchers and data scientists ..." We don't legitimize NSA that way, so why do we legitimize it for Facebook? Does something become morally acceptable just because it is done for money?

girvo• 13 years ago

Difference is, Facebook can't send me to jail. Or kill me.

wcoenen• 13 years ago

The combination of AI and social networks reminds me of Friendship is Optimal[1].

It's a story about an AI who has the job "to satisfy everybody's values through friendship and ponies" in an MMORPG but breaks out and starts optimizing the real world.

[1] http://www.fimfiction.net/story/62074/

Facebook's AI lab

Related Stories